PretiMeth

PretiMeth predicts continuous DNA methylation levels at single-CpG resolution from array-derived methylation features to enable locus-level methylation inference and downstream differential-methylation analyses.

Key Features:

• Single-CpG resolution prediction: Provides continuous methylation predictions at individual CpG loci, enabling interrogation beyond conventional array coverage.
• Locus-specific models: Trains an independent prediction model for each CpG locus using logistic regression to allow per-locus performance assessment.
• Single-feature selection: Selects a single DNA methylation feature most similar to the target CpG site to parameterize each model.
• Robust performance: Demonstrates improved prediction accuracy relative to alternative algorithms with consistent performance across datasets and biological contexts.
• Application to large cohorts: Applied to large datasets (e.g., TCGA) to identify differentially methylated CpG loci and genes between tumor and normal samples.
• Probe-set design utility: Uses high-precision, minimal-feature models to inform design and optimization of DNA methylation beadchip probe sets.

Scientific Applications:

• Locus-level methylation inference: Extends the utility of methylation array data to infer methylation at individual CpG loci.
• Differential methylation analysis: Identifies differentially methylated CpG loci and genes in large cohorts such as TCGA between tumor and normal samples.
• Probe design and optimization: Informs design and optimization of updated DNA methylation beadchip probe sets using minimal predictive features.
• Biomarker and target discovery: Supports biomarker discovery, candidate therapeutic target prioritization, and studies in personalized medicine and molecular biology.

Methodology:

Trains per-CpG logistic regression models that use a single highly correlated DNA methylation feature selected for each target locus as the model predictor.